.
Finally, what tests should I run for “yuva16”?
Try to render 10bit test image to various image/ffmpeg formats, see if all bits still there (by subtracting rendered output and original), may be use longer 10bit video to see if yuva16 offer any speed up on your machine compared to rgba-float.
Try some plugins. I think unsupported plugin will just give you black frame in compositor, but not sure.
Did Andrew or Andrea get any bad results with yuva16? should we add this patch to the source? it is just suspicious because the lines before were commented out for some reason, which I assume is because it behaved badly.
I think they behaved badly in sense only SOME plugins supported them, and 32 bpc floating point was superior in image-processing/compositing.
Thing is, most video codes even today are integer-based in their input and output, so while ffmpeg can do 10 bit yuv directly to another encoder input we go via 444 yuv with 16 bpc, THEN via twice wider fp format, then back to 16 bpc 444 yuv, then into that codec (encoder) supports. Quite a lot of conversions!
I'm quite happy they were NOT removed, because now we can compare them to other formats.